Intelligent tensor product mode transformation-based three-sliding-surface sliding mode controller design

Abstract

This paper studied a novel chattering elimination method based on improved three-sliding-surface and tensor product mode transformation, and the intelligent adaptive sliding mode controller with reinforcement learning strategy is proposed for parallel-type double inverted pendulum (PIP) system. The tensor product model transformation based sliding mode controller was introduced firstly to lay the basis for intelligent adaptive sliding mode controller, and then an adaptive reinforcement learning algorithm is utilized to facilitate controller's performance. In this paper, reinforcement learning algorithm is employed to find the instantaneous optimal value for the boundary layer width of saturation function which is important in the controller chattering reduction. The proposed controller with reinforcement learning strategy is verified by PIP system whereas the agent is rewarded for lower chattering situation and punished for higher case. Simulation results show that chattering can be reduced effectively by the adaptive three-sliding-surface sliding mode controller.